The imu (raw) has a bias in the x component

[mllofriu]

Please provide the following information:

• OS: Ubuntu 20.04 in Tegra
• ROS Distro: Foxy
• Built from source or installed: Installed, built jackal packages from source
• Package version: ee4e3f6

Expected behaviour
The messages in /imu/data_raw have a near zero values in the x and y components when the robot is still.

Actual behaviour
A bias observed throughout a log, in which the robot starts at a still position.

To Reproduce
Provide the steps to reproduce:

1. run jackal bringup
2. let the robot be still
3. ros2 topic echo /imu/data_raw --csv
4. plot with R

require('reshape2')

read_csv_and_groom <-function(path, cols_to_select, source) 
{
  data <- read.csv(path, header = FALSE)
  data <- data[,cols_to_select]
  colnames(data) <- c('ts', 'ns','x', 'y', 'z')
  data$ns <- as.integer(data$ns)
  data$ts <- as.integer(data$ts)
  # transform secs rel to start
  data$ts <- data$ts - data$ts[1]
  # keep only milliseconds
  data$ms <- data$ns / 1e6
  print(data$ns)
  print(data$ms)
  data$ms <- data$ts * 1e3 + data$ms

  data <- melt(data, na.rm = TRUE, id.vars = 'ms', measure.vars = c('x', 'y', 'z'))
  data$source <- source
  
  data
}

imu <- read_csv_and_groom('~/Downloads/imu.csv', c(1,2,29,30,31), 'imu')
mag <- read_csv_and_groom('~/Downloads/mag.csv', c(1,2,4,5,6), 'mag')

all <- rbind(imu, mag)

require('ggplot2')
ggplot(all[all$source == 'imu',], aes(x = ms)) +
  geom_point(aes(y = value, color=interaction(source, variable))) 

[tonybaltovski]

The accelerometer when stationary is measuring the force of gravity. If the platform is not perfectly aligned with Earth's gravity vector it could have X or Y direction offsets. Is the Jackal started on a flat surface? Does the filtered data have the same offset?

[mllofriu]

The jackal is started in a flat surface. It's just running on the floor.

The z component seems close to the expected value for the gravity, so I wouldn't expect a significant component in x or y (unless that is not to scale).

By 'filtered output' do you mean the madgwick filter? Does that correct for a potential offset of the imu_link?

[marco-ambrosio]

I would like to add that the data published on /Imu/data_raw are without covariance, thus standard filter implemented in ROS can do pretty nothing on them